One of the things that has been lacking with Nvidia Grid vGPU is the ability to monitor vGPU performance. Up until now this could only be done from host level, and only framebuffer at VM level. I’m very happy that Nvidia has listened to customers and community and has now greatly improved vGPU monitoring. With the August 2016 release of GRID you can now monitor a lot of new metrics in a lot of new ways.
This means that it will be easier both from host level and VM level to understand what is going on, who is using the resources and where is the bottleneck.
Cool thing is that you can use existing tools like Perfmon, Nvidia-SMI and WMI to read the performance metrics. They also worked with all the big vendors to get the metrics into existing monitoring solutions. So you don’t need to buy a new monitoring solution you can most likely use the existing one or build your own tools.
Why is this so important? Because performance monitoring is key to success at all levels of GPU virtualization projects. Now you can assess exiting workstations performance, size you GPU environment and then optimize it after implementing it. Then you will always be able to get the best user experience at the highest possible density and lowest cost.
I’m very excited about this because I’m developing my own tools for assessment and monitoring vGPU, so I’m very happy to say that I will release two new tools very soon that supports the new vGPU monitoring:
GPUPerf 3.0 and GPUSizer 1.0
This is a follow up of a tool I made over a few years for realtime performance and protocol data. In the new version you will see vGPU performance, how much GPU and CPU is used for encoding. There is also support for new protocols like VMWare Blast and Citrix Framehawk.
The tool already works with vGPU utlilization, but I’m also implementing more of the new metrics like encoding utilization. Need to test this first.
I will release this tool as soon as possible, most likely after vmworld. Follow my blog or me on twitter : @magnarjohnsen for updates.
GPU sizer is a tool you can use to assess and optimize a vGPU implementation. You run it on your workstations and monitor you GPU usage. Then it will in real-time suggest the right Nvidia Grid card and vGPU profile for you. After implementation you can use the tool to verify that you have the right vGPU profile and you can resize based on analytics provided by the tool. This tool makes it extremely easy to implement, optimize and monitor vGPU solutions. In the picture below you can see a youtube video running on XenApp with K2 passthrough mode. You can quickly see that by using two new M10 boards you could have up to 128 users like this.
Please sign up here if you are interested in this tool, currently only in Beta version and should work with vGPU already.